The Nvidia GeForce RTX 5080 Founders Edition review: monster AI gaming from $999

Rather than racing out the door to capitalise on the breakneck momentum set by Blackwell’s flagship, Nvidia GeForce RTX 5080 strolls out leisurely, aware it’s uncontested. There’s a calm air about it, knowing nothing out there hits the same beat and won’t for quite some time.

All eyes here at Club386 continue to wander over to the mighty GeForce RTX 5090 with dreams of sky-high performance, but there’s enough here to keep your attention at half the price. It doesn’t break the bank in the same way at $999, matching the cost of RTX 4080 Super at launch. Whether that keeps is another question entirely, but it’s off to a good start.

Blackwell architecture brings with it solutions to concerns of stagnating rasterised performance, leveraging Nvidia’s AI expertise to trade brute force for much-improved neural rendering techniques. GeForce RTX 5080 is our first indicator of just how well this sorcery scales down to the high-end and indeed gives us a glimpse at what we can expect from future entries to the series.

Let’s consult with the trusty Club386 Table of Doom™ to see where it sits compared to predecessors.

Specifications

GeForce RTX	5080	4080 Super	3080
Launch date	Jan 2025	Jan 2024	Sep 2020
Codename	GB203	AD103	GA102
Architecture	Blackwell	Ada Lovelace	Ampere
Process	TSMC 4N (5nm)	TSMC 4N (5nm)	Samsung 8N
Transistors (bn)	45.6	45.9	28.3
Die size (mm2)	378	378.6	628.4
SMs	84 of 84	80 of 80	68 of 84
CUDA cores	10,752	10,240	8,704
Boost clock (MHz)	2,617	2,550	1,710
Peak FP32 TFLOPS	56.3	52.2	29.77
RT cores	84 (4th Gen)	80 (3rd Gen)	68 (2nd Gen)
RT TFLOPS	170.6	121	58
Tensor cores	336 (5th Gen)	320 (4th Gen)	272 (3rd Gen)
FP16 Acc TFLOPS	450.2	418	238.2
Peak FP4 TFLOPS	1,801	836	–
ROPs	112	112	96
Texture units	336	320	272
Memory size (GB)	16	16	10
Memory type	GDDR7	GDDR6X	GDDR6X
Memory bus (bits)	256	256	320
Memory clock (Gb/s)	30	23	19
Bandwidth (GB/s)	960	737	760
L2 cache (KB)	65,536	65,536	5,120
PCIe interface	Gen 5	Gen 4	Gen 4
Video engines	2 x NVENC (9th Gen) 2 x NVDEC (6th Gen)	2 x NVENC (8th Gen) 1 x NVDEC (5th Gen)	1 x NVENC (7th Gen) 1 x NVDEC (5th Gen)
Power (watts)	360W	320W	320W
MSRP	$999	$999	$699

GeForce RTX 5080 uses all the GB203 chip has to offer, packing it to the brim. All seven graphics processing clusters (GPCs) are in effect, giving it a modest 5% more CUDA cores than its predecessor at 10,752. That’s half of what you get inside the top-of-the-range GeForce RTX 5090, but still a decent leap from where we were just four years ago.

Some might question Nvidia steering the course with an identical memory configuration as GeForce RTX 4080. HD texture packs and ray tracing raise the ceiling ever higher, potentially leaving 16GB across a 256-bit bus something of a stickler in some games. As the old adage goes, it’s not how big it is, but how you use it.

Nvidia has moved to high-speed GDDR7, amping up speeds courtesy of 30Gb/s modules – the fastest we’ve seen yet. There is a desire for more as the total bandwidth is once again half its bigger brother at 960GB/s, but that’s still a 30% leap over other 80-class cards.

Of course, none of this paints the full picture of Blackwell. Nvidia kicks off its AI-led revolution with some impressive architectural tinkering and AI witchcraft in an attempt to create a card that’s great for both work and play.

Architecture

GeForce RTX 5080 elicits some deja vu with hardware, featuring the same TSMC 4N process as its predecessor. With an ever-so-slightly smaller 378mm² die, Nvidia carves out room for its expanded 84 streaming multiprocessors (SMs) by dialling back transistors a touch. A definitively worthwhile trade, owing to its new shader layout packing a huge punch.

Each SM carries the same general configuration of 128 shaders, four Tensor Cores, four Texture Units, and a single RT Core, but with a couple of key differences. Firstly, Tensor Cores enter their fifth generation, now supporting new quantisation formats and precisions to speed up training and utilising machine learning. It’s Nvidia’s AI bedrock decades in the making.

Secondly, and more importantly for a gaming card such as this, cores can now process either floating point or integer workloads. We go into more detail in our GeForce RTX 5090 review, but this essentially leaves no wasted potential as games make use of INT32. Being even handed, Nvidia doesn’t increase the ROP count between generations, meaning raster performance won’t see the jump it could have done.

Multi-frame Generation

Nvidia came to play with its RTX 40 Series, boosting performance with Frame Generation on top of AI upscalers. The premise is simple enough, using Tensor Cores to sandwich a single AI-generated frame in between each rasterised frame for a sweet 55% fps jump using DLSS Quality in our GeForce RTX 4090 review.

Taking things a step further, Blackwell amplifies this threefold with its new Multi-frame Generation (MFG) slipping in a total of three synthetic frames for every rendered one. You can see the results for yourself by heading into the performance section, but in many cases, it can net up to double the frame rate in some games.

It’s an idea Nvidia has long ruminated on but only managed to implement successfully into RTX 50 Series after a few necessary tweaks under the hood. Not quite the full-on Google approach to software, instead creating harmony with the hardware underneath.

Improved Flip Metering nails the timing of each AI-generated frame, keeping them in line with rasterised ones to prevent stumbling and stuttering that would otherwise make all improvements null and void. Also gone is the previous hardware-based Optical Flow Accelerator, traded in for a more reliable deploy-once AI model that’s both 40% faster and consumes 30% less VRAM.

16GB of GDDR7 on top of RTX 5080 starts to look a little more understandable as a result. Finally, a new AI Management Processor (AMP) keeps all of this in check like a good project manager should.

Transformer Model

Part and parcel with fresh rendering technology is a new AI model it can use to generate pixels. Convolutional Neural Networks (CNN) is still alive and kicking in games that support super resolution, but Nvidia has already refined it to within an inch of its life. Whatever pitfalls you’ve noticed when using it, such as ghosting and blooming, are there to stay.

The only way to improve visual fidelity and combat those pesky artefacts is to switch things up, which is where Google’s Transformer Model comes in. Dating back to 2017, it’s currently the most popular model seen within ChatGPT and Gemini, and for good reason. It carries a larger overhead, as Nvidia claims it requires twice the parameters and four times the compute compared to CNN, but with it comes improved temporal stability that’s the closest we’ve ever been to native without directly rendering each individual frame.

RTX 5080 and its Blackwell cohorts benefit most with the enhancements it brings to Multi Frame Generation, but it’s not confined to the latest lineup. It also improves standard Frame Generation for RTX 40 Series cards, and extends its reach to uplift Ray Reconstruction, Super Resolution, and Deep Learning Anti-Aliasing (DLAA) across the entire RTX ecosystem. You can run an RTX 2060 and you’ll still enjoy the augmentation.

The only caveat is one it shares with MFG in that it requires developer implementation. DLSS Override from within the Nvidia App only takes you so far without the codebase to work from. I have my highest hopes that devs will be quick to set it all up after seeing the wonderous things it can do in Alan Wake 2 and Cyberpunk 2077, but that’s the test of time as we know it.

Founders Edition

Nvidia has the craft of its Founders Edition models down to a fine art at this point, producing some of the sexiest graphics cards on the market. If you ask me, RTX 40 Series lost the magic a little as its three-slot bulk didn’t suit the professional aesthetic Team Green strives for, but it was a necessary evil of sorts to tame the beast beneath. Fortunately, elegance is well and truly back in RTX 50 Series thanks to major cooling innovations.

Returning to a slimmer two-slot form factor, GeForce RTX 5080 has no trouble fitting in most rigs at 304mm x 137mm. These dimensions qualify it for Nvidia’s SFF-ready badge of honour, although you’re always best whipping out your trusty tape measure before sticking it inside a mini PC.

Sharing the exact same chassis as GeForce RTX 5090 FE, Nvidia toes a familiar line with its colour scheme. As much as I’d love to see a bleached Founders Edition primed for my white build, there’s no arguing that grey and black fit in most themes. Gone is large, front-facing branding, replaced with a subtle etching to let everyone know which model you sport, while dinky LED strips accentuate the eye-catching X pattern in the centre.

Both fans now sit side by side in a more traditional arrangement, blowing air past the new vapour chamber and out through two concaved fin stacks. The combination of curved and straight fins helps reduce pressure, with the longest optimising airflow where it’s greatest. Meanwhile, angled exhausts on the side do their darndest to direct heat away from the card rather than recycling it back through the fans again.

All in all, it’s a lot of engineering to keep the uniquely centralised 378.6mm² die cool on top of Nvidia’s smallest PCB to date, but the changes don’t stop there. Relocating everything to the middle of the card increases the distance between the board and both the PCIe slot and I/O, prompting boffins to spin a web of proprietary PCBs and ribbon cables to reach all the necessary ports.

None of this technical wizardry slows RTX 5080 down in the slightest. In fact, Nvidia outfits it with the latest cables out there, and its single HDMI 2.1b and three DisplayPort 2.1b slots have enough bandwidth to push up to 4K 12-bit HDR at 480Hz or 8K 12-bit HDR at 165Hz – alongside display stream compression (DSC) that is.

The return of the 12VHPWR power connector is no surprise despite its exceptionally rocky start in life. After all, its 16 pins deliver an ample 600W current through a single cable, which is more than enough to saturate RTX 5080’s 360W hunger. You’ll be pleased to know Nvidia has made some tweaks to the interface by following the much-improved 12V-2×6 variant, reinforcing the contact block and securing the pins. Still, if you’re a worry wart like me, MSI adapters (sold separately) offer peace of mind with yellow housing to ensure nothing’s loose.

I highly suggest pairing RTX 5080 with a modern power supply outfitted with a 12VHPWR header, as it offers a streamlined look with one thin cable from end to end. However, you don’t need to invest in a new PSU just for a simple wire. You’ll spot a three-way, eight-pin splitter in the box that’ll hook up to most units out there. Just keep in mind Nvidia recommends 850W to account for all your other components.

Better safe than sorry, but 850W seems like a conservative estimate to me. RTX 5080 is a rather polite guest inside our beefy Club386 test bench, as the whole system runs at just 475W when gaming and 101W while idling.

This is less than Radeon RX 7900 XTX, which itself carries a slimmer 800W suggestion from AMD. Fill out four memory lanes, perhaps add in an RGB strip or two and power expectations will creep up, but the ceiling seems lower than initially anticipated, which is a welcome bonus.

Slimming power demands within 4% of its predecessor pairs well with Nvidia’s Double Flow Through cooler, keeping RTX 5080 as cool as a cucumber – you must like hot cucumbers (ed). Peak gaming temperatures reach no higher than 64°C, which is impeccable for a dual-slot solution eschewing liquid metal in favour of a traditional thermal compound solution, all while fans spin at just 39.5dB. We’ve yet to hear anything quieter, which is quite the oxymoron.

A quick reminder of its $999 price tag, and GeForce RTX 5080 appears to be one of the most compelling cards in the Blackwell line-up. Sadly, you might struggle to get your hands on one if rumblings of stock shortages are anything to go by, particularly in a sleek Founders Edition dress. Just keep an eye on Nvidia’s website and your finger hovering over F5 any time from tomorrow, January 30.

Performance

To capture where GeForce RTX 5080 sits in the current market, we’ve tested a total of nine graphics cards from scratch. Spoiler alert: armed with the latest drivers in their respective ecosystems, even the best AMD Radeon struggles to earn its place against Nvidia’s latest high-end powerhouse.

Application & AI

Leaked 3DMark Time Spy results already spoiled this one for us ahead of launch as other tests in the Futuremark suite seem to follow suit. RTX 5080 shows its ray tracing chops in 3DMark Speed Way by slotting in between two predecessors, offering a 20% uptick over RTX 4080 Super while just 14% shy of RTX 4090.

It’s no match for the mighty RTX 5090, but 37% less performance doesn’t seem so bad when you save half the cost. Meanwhile, AMD Radeon RX 7900 XTX just about claws into the top-five graphics cards we’ve tested by default. One has to wonder if that’ll stay the case when more Blackwell GPUs arrive.

3DMark Steel Nomad strips all the fancy rendering techniques to focus purely on rasterisation, yielding fairly similar results to its ray-traced counterpart. RTX 5080 once again falls behind the flagship’s staggering lead by 40%, but it still over performs relative to its price – and this is before we get into the magic of AI.

It’s a far closer fight in the high end, as RTX 5080 tickles the heels of RTX 4090 with just 11% in it between the two. Otherwise, you can expect anything from a 23% lead over its rivals and predecessors.

RTX 5080 might be more of a gaming card, but content creation isn’t lost on the pusher of pixels. It’s capable of 9,189 samples per minute in Blender, placing it 39% short of its bigger brother and 18% behind the Ada Lovelace champion. Again, it’s hard not to highlight the disparity in value here, given it also saves you 50% and 33% from their respective MSRPs.

Comparatively, RTX 5080 is just 8% more productive in 3D rendering than the likes of RTX 4080 Super, but for the same cost. One for one, it’s a net positive. This is sure to perk up the ears of RTX 30 Series or AMD Radeon owners, who net anything from 64% to double the performance.

Given Nvidia’s focus on artificial intelligence, it’s no surprise to see RTX 5080 standing tall as the only graphics card able to go toe-to-toe with the king. It’s still 16% short of RTX 5090 in Geekbench AI, but a comfortable 15% ahead of the closest competition. As far as cost-conscious AI performance goes, there is simply nothing else in this price range.

Leaving the flagship to play in a league of its own, RTX 5080 is almost neck and neck with RTX 4090 in Procyon’s AI Text Generation benchmark. This subjects our graphics cards to several large language models (LLMs) using Llama 3.1 as a framework.

Once again expect to see double the performance of AMD Radeon’s best and brightest, a minimum 31% leap from Nvidia’s Ampere Series, and a 19% jump from the previous generation. It’s safe to say it’ll make short work of generative AI.

Game Rasterisation

Synthetic tests tell but half the story, so allow good old games to fill in the real-world blanks. Our baseball team of graphics cards is batting against a roster of triple A titles, each pitching with the three most popular resolutions – 1920×1080 (FHD), 2560×1440 (QHD) and 3840×2160 (UHD).

Credit where it’s due, AMD Radeon RX 7900 XTX shows up when it comes to crunch time, but it’s still not enough to topple RTX 5080 in Assassin’s Creed Mirage. Granted, Team Red might clinch the value proposition with just a few frames in it between the two, but don’t count out Blackwell’s high-end just yet as we’ll soon see its secret weapon in action.

Speaking of price to performance, RTX 5080 sits within just 12% of the flagship at FHD. This grows to 20% at QHD and 27% at UHD, but still half the price you’ll pay for the pleasure. It’s starting to look like the card gamers dream about.

It’s a similar story with Final Fantasy XIV: Dawntrail. RTX 5080 proves its mettle in FHD compared to 90-class alternatives but still keeps a respectable margin at higher resolutions. It outpaces everything AMD has to offer, while falling just 35% short of the top dog.

You might get more out of RTX 4090 if you hunt around for a good deal, but it’ll need to be lower than $700 to offer anything of substance. All in all, it just further underscores how much of a bargain RTX 5080 is at MSRP.

Forza Motorsport is a particularly demanding game where even the best graphics card in the world can’t achieve triple-digit frame rates at 4K. Fortunately, RTX 5080 keeps a smooth frame rate at all resolutions compared to most of its peers, proving a worthwhile upgrade from Ampere and Radeon 7000 Series alike. It even steals some thunder from RTX 4090, sitting withing just 7fps.

Mount & Blade II: Bannerlord is the first time we see RTX 5080 wobble, slipping halfway down the pack. It’s hard to whine about any graphics card capable of 350fps, but there’s no denying that the strategy game responds to other models better at lower resolutions.

Crank things higher and the Blackwell GPU starts to come into its own again, leaping 17% past its predecessor and 6% over its rival at UHD.

Slight fluctuations in Tom Clancy’s Rainbow Six Extraction don’t detract from the overall theme that RTX 5080 belongs in between RTX 4080 Super and RTX 4090 in terms of rasterised performance, regardless of resolution. Simply take a look at this averaged table to see for yourself using UHD as a reference point:

	Average increase or decrease at 4K
RTX 5090 to RTX 5080	-32%
RTX 4090 to RTX 5080	-15%
RTX 4080 Super to RTX 5080	+14%
RX 7900 XTX to RTX 5080	+15%
RX 7900 XT to RTX 5080	+37%
RTX 3090 to RTX 5080	+52%
RTX 3080 to RTX 5080	+77%
RX 7800 XT to RTX 5080	+84%

Moving to Blackwell might prove a tough sell for previous generation owners going off rasterised performance alone, but RTX 5080 is a far more meaningful option for anyone coming from the midrange or something a little older. Still, let’s not draw any conclusions just yet. There’s still the matter of all those AI accelerators under the hood.

Frame Generation

First, a quick history lesson about frame generation (FG) to lay the groundwork. Call of Duty: Black Ops 6 benefits from both FSR and DLSS, but our RTX 5080 barely sees a bump moving from native performance to Quality upscaling. Instead, the largest uplift comes from added AI-generated frames, seeing 30% more than other rendering techniques.

This is true of all models, with DLSS Frame Generation coming out slightly ahead of FSR3 Frame Generation for supported RTX 40 and 50 Series GPUs. To paraphrase Happy Gilmore, “it ain’t over yet. The way Nvidia sees it, we’ve only just begun.”

Putting all of Nvidia’s upscalers against the wall, 3DMark’s new DLSS test shows a decent uplift simply moving from one generation of FG to the next. Alongside reflecting on just how far we’ve come from the small beginnings of DLSS 2, there’s proof in the pudding that even RTX 40 Series benefits from Team Green’s tweaking.

It doesn’t stop there, though. GeForce RTX 5080 uses Multi Frame Generation, which slips a second (3x) and third (4x) AI-generated frame in for good measure. While our high-end card already performs admirably, it eclipses RTX 4090’s capabilities by up to 55% when the feature is dialled up to 11, cementing its place as the second-best graphics card money can buy right now.

Putting this into practice, an early build of Alan Wake 2’s MFG update shows just how much the survival horror game benefits from the new feature. Settings such as path tracing make even the strongest graphics cards buckle without a helping hand, but RTX 5080 goes from barely playable at UHD using just DLSS to maxing out 144Hz gaming monitors with room to spare. It feels like magic, considering it’s all nestled in a single option.

Much like Nvidia showcased during its reveal, it has noticeably polished its neural rendering suite, as there are far fewer shimmers, glimmers, ghosting, and blooming than previous upscalers. However, that doesn’t mean MFG is perfect. Keen eyes will eventually spot the odd thing that doesn’t belong, but to my peepers, neon signs boast more clarity than ever, and finer details like wires and hair behave as they should.

While pure rasterisation debatably feels better as you pan the camera around the titular character, 153fps with MFG is undoubtedly smoother than sub-60fps without the artificial leg up. Synthetic is simply the way forward as brute force hardware yields diminishing returns. Nvidia is just refining it quicker than anyone else.

Cyberpunk 2077 is another one renowned, or perhaps infamous, for its incredibly demanding ray-tracing presets. A graphics card even remotely glancing at Overdrive starts to work up a sweat. You can see for yourself in the tables above, with Radeon struggling to muster up enough gumption to break triple digits at FHD and falling far short of the 60fps ideal at UHD.

Multi Frame Generation changes the game entirely, delivering comfortable frame rates across the board. It works best with a base of at least 60fps to work with, but that doesn’t stop 4x frame generation from boosting RTX 5080’s lowly 19fps native showing at 4K to a staggering 137fps – a 621% increase.

Of course, it’s early days for MFG and I’ve yet to see how it plays out across the 75 announced launch titles, but it makes a solid first impression. Its scalability opens plenty of doors, from higher resolutions such as 8K when displays inevitably land on our doorstep to competitive frame rates without giving up that all important detail, and improved battery life should it ever grace a gaming handheld.

As much as my mind can wander musing the endless possibilities of its bright future, I can’t argue with its impact in the here and now, too. Just make sure you check DLSS MFG support in your favourite titles before deciding to cop an RTX 50 Series card for yourself, as while there’s possibility of using it via override in the control panel, optimised implementation remains in the hands of developers.

Conclusion

With GeForce RTX 5080, Nvidia cements Blackwell graphics cards as the go-to for gamers, content creators, and AI enthusiasts. Much like GeForce RTX 5090, it arrives on the scene with no equal yet carries a slimmer $999 price tag than RTX 4080 to lower the barrier to entry. No small chunk of change, but those with deep enough pockets will find value, even compared to the flagship. It’s half the price, but always more than half its performance.

Leaner generational gains in rasterised performance alongside sticking with 16GB of memory make it less enticing for anyone jumping ship from high-end RTX 40 Series solutions, but Nvidia has built on its frame-generation foundations well. Combined with a smarter neural network, RTX 5080 pushes supported games to the nth degree, often doubling anything RTX 4080 Super can handle while sporting fewer artefacts in the process.

Optimised Multi Frame Generation compatibility rests in the hands of developers, making additional performance something of a bonus rather than a given. Still, 75 games and apps right off the bat shows commitment to the cause, instilling faith that there are many more to come. Even without the promise the feature will appear in your favourite title, RTX 5080 is a fine card.